Assessing replication success via skeptical mixture priors